Positivstellensätze for algebras of matrices (Q649579)

Let \(\mathcal{A}\) be an algebra on which a meaningful notion of positivity exists. A `positivstellensatz' for \(\mathcal{A}\) is a theorem saying that elements in \(\mathcal{A}\) that are positive have an algebraic description turning positivity `evident'. The classical example is Artin's solution to Hilbert's 17th problem according to which a rational function on \(\mathbb{R}^n\) that is nonnegative is a sum of squares of such functions. Positivstellensätze for commutative algebras have meanwhile a rich history that is reported in the books by Bochnak, Coste, Roy; Delzell and Prestel; and Marshall. More recently, positivstellensätze for noncommutative algebras have been proved. At the beginning of this endeavour we have \textit{J. W. Helton} [Ann. Math. (2) 156, No. 2, 675--694 (2002; Zbl 1033.12001)]. In the present paper, positivstellensätze are proved for path algebras, matrices over certain noncommutative and commutative algebras, crossed product algebras, matrix polynomials in intervals, matrices over fields, and cyclic algebras. Sections 1 and 2 give much of the necessary definitions and notations fundamental for what follows: Let \(\mathcal{A}\) be a \(*\)-algebra, i.e., an algebra over \({\mathsf L}={\mathsf R}\) or \({\mathsf C}\) with involution, where \({\mathsf R}\) is a real field and \({\mathsf C}={\mathsf R}(\sqrt{-1}).\) If \((\mathcal{V},\langle \cdot,\cdot \rangle)\) is a unitary space (over \(\mathbb{C}\)), let \(\mathcal{L(V)}\) be the algebra of linear mappings of \(\mathcal{V}\) and let \(\mathcal{L}^+(\mathcal{V})\) denote the largest \(O^*\)-subalgebra; i.e., the largest subalgebra of \(\mathcal{L (V)}\) with the property that for each of its elements \(A\) there exists an adjoint \(A^*\) in it -- i.e., \(\forall v,w \in \mathcal{V}\) we have \(\langle Av,w\rangle= \langle v,A^*w\rangle.\) Recall that the standard notion of positivity on \(\mathcal{L(V)}\) is given by \(A>0\) iff \(\langle Av,v\rangle>0\) for \(v\neq 0.\) A \(*\)-representation of a \(*\)-algebra \(\mathcal{A}\) over \(\mathbb{C}\) is a homomorphism \(\pi:\mathcal{A} \rightarrow \mathcal{L^+(V)}.\) Let \(\mathcal{A}_h=\{a\in \mathcal{A}:a^*=a\}\) (with \(\mathcal{A}\) over \({\mathsf L}\)) be the self adjoint or Hermitian elements of \(\mathcal{A}\), and \(\mathcal{A}^\circ=\{x\in \mathcal{A}\setminus \{0\}: x \text{ is not zero divisor}\}.\) A set \(\mathcal{C}\subseteq \mathcal{A}_h\) is a quadratic module if it is closed under nonnegative linear combinations, contains 1, and satisfies \(x^*\mathcal{C}x\subseteq \mathcal{C}\) for all \(x\in \mathcal{A}.\) The smallest quadratic module in \(\mathcal{A},\) denoted \(\sum \mathcal{A}^2,\) is the set of sums of form \(x_1^*x_1+\cdots+x_m^*x_m\) with \(x_i\in \mathcal{A}.\) A quadratic module \(\mathcal{C}\) defines a noncommutative preordering if the product of any two commuting elements in \(\mathcal{C}\) belongs to \(\mathcal{C}.\) Denote the smallest noncommutative preordering by \(\sum_{ nc} \mathcal{A}^2.\) Given \(a\in \mathcal{A},\) a set \(\mathcal{S}_a \subseteq \mathcal{A}_h\) containing \(a,\) so that for all \(x\in \mathcal{A},\) \(x^*\mathcal{S}_a x \subseteq \mathcal{S}_a;\) and \(bc=cb,~ b\in \mathcal{S}_a,~ c\in \sum_{nc} \mathcal{A}^2\) together imply \(bc\in \mathcal{S}_a\) is a denominator set for \(a.\) Finally, given a \(*\)-algebra \(\mathcal{A}\) over \({\mathsf L}\) with \(*\)-subalgebra \(\mathcal{B},\) certain linear and involution-preserving surjections \(p:\mathcal{A} \rightarrow \mathcal{B}\) so that \(p(\sum \mathcal{A}^2) \subseteq \sum \mathcal{B}^2\) are called strong conditional expectations. With these definitions and a notion of positivity on \(\mathcal{A},\) let \(\mathcal{A}_+\) be the positive elements in \(\mathcal{A}.\) A positivstellensatz of type I, II, III, IV, V is satisfied by \(x\in \mathcal{A}_+\) according to the following: type I: if \(x\in \sum \mathcal{A}^2;\) type II: if there exists a \(c\in \mathcal{A}^\circ\) so that \(c^*xc \in \sum \mathcal{A}^2;\) type III: if there is a \(c\in \mathcal{A}^\circ \cap \sum \mathcal{A}^2\) so that \(xc=cx\) and \(xc \in \sum \mathcal{A}^2;\) type IV: if \(x\in \sum_{nc} \mathcal{A}^2;\) type V: if there is \(c_x\in \mathcal{S}_x\) so that \(\sum_{nc} \mathcal{A}^2.\) Algebra \(\mathcal{A}\) is said to satisfy a positivstellensatz of type X if all its positive elements satisfy the respective positivstellensatz. Section 3 recalls the notion of the path algebra \(\mathbb{C} \Gamma\), defined in a natural way from a directed multigraph \(\Gamma\) that contains with each arrow \(b\) also its reverse \(b^*.\) Let \(\mathcal{F}=\mathbb{C}\langle a_1,\dots,a_m,a_1^*,\dots,a_m^* \rangle\) be the free \(*\)-algebra over \(\mathbb{C}\) on \(m\) generators. In Proposition 3.1, the authors generalize the main theorem of Helton [loc. cit.]: a Hermitian element in the matrix algebra \(M_n(\mathcal{F})\) satisfies \(\rho(X)\geq 0\) for every finite dimensional representation \(\rho,\) if and only if \(X\in \sum M_n(\mathcal{F})^2.\) They then define a strong conditional expectation from \(M_n(\mathcal{F})\) onto \(\mathbb{C} \Gamma\) to prove a similar type I positivstellensatz for the path algebra in a manner different from \textit{S. Popovych} [J. Algebra 324, No. 9, 2418--2431 (2010; Zbl 1219.46068)]: a Hermitian \(X\) in \(\mathbb{C} \Gamma\) allows only nonnegative images under finite dimensional \(*\)-representations if and only if \(X=\sum_{j=0}^k X_j^* X_j\) for some \(X_j\in \mathbb{C}\Gamma.\) Section 4 proves an Artin theorem for matrices over noncommutative zero divisor free unital \(*\)-algebras that satisfy the left Ore condition: given \(a\in \mathcal{A}, s\in \mathcal{A}^\circ=\mathcal{A}\setminus\{0\},\) there exist \(b\in \mathcal{A}, t\in \mathcal{A}^\circ\) so that \(ta=bs.\) First diagonalizability results for Hermitian matrices over \(\mathcal{A}\) are shown. Next suppose \(\mathcal{A}\) to be an \(O^*\)-algebra on a unitary space. Then natural definitions of positivity yield \(\mathcal{A}_+\) and \(M_n(\mathcal{A})_+.\) Theorem 4.8 shows that a type II positivstellensatz holds for the matrix algebra \(M_n(\mathcal{A})\) provided it holds for \(\mathcal{A}\) itself. In Section 5 for a unital \(*\)-algebra \(\mathcal{A},\) and a finite group \(G\) of \(*\)-automorphisms of \(\mathcal{A},\) the crossed product \(\mathcal{A} \times_\alpha G\) is defined -- see \textit{R. S. Pierce} [Associative algebras. New York-Heidelberg-Berlin: Springer (1982; Zbl 0497.16001)]. One then has the regular covariant representation \(\pi_{creg}:\mathcal{A} \times_\alpha G \rightarrow \mathcal{L}(\oplus_{g\in G} \mathcal{V})\) which defines in analogy to earlier observations a notion of positivity via \((\mathcal{A} \times_\alpha G)_+=\{x\in \mathcal{A} \times_\alpha G: \pi_{creg}(x) \geq 0\}.\) Then if a positivstellensatz of type I or type III holds for \(M_n(\mathcal{A})\) then - again via a strong conditional expectation -- the respective theorem will hold for \(\mathcal{A} \times_\alpha G.\) Also if \(\mathcal{A}\) is an \(O^*\)-algebra without zero divisors and \(\mathcal{A}^\circ\) is left Ore, we have: if a type II satz holds for \(\mathcal{A}\) then it holds for \(\mathcal{A} \times_\alpha G.\) Section 6 assumes \(\mathcal{A}\) to be a finitely generated unital and commutative \(*\)-algebra without zero divisors over \(\mathbb{R}\) or \(\mathbb{C}.\) By \(\hat{\mathcal{A}}\) the set of characters is understood and it is assumed that \(\sum_j a_j^*a_j=0\) implies \(a_1=a_2= \cdots=a_n=0.\) This can be guaranteed if \(\hat{\mathcal{A}}\) separates points. Positivity notions are given by \(\mathcal{A}_+ =\{a\in \mathcal{A}: \chi(a)\geq 0 \text{ for } \chi \in \hat{\mathcal{A}}\}\) and \(M_n(\mathcal{A})_+=\{A: (\chi(a_{ij}))\geq 0 \text{ for }\chi \in \hat{\mathcal{A}}\}.\) Then a type II positivstellensatz holds for \(M_n(\mathcal{A}).\) The proof uses induction on \(n\) and diagonalization. Section 7 recalls a matrix version of the classical Fejér-Riesz theorem due to \textit{M. Rosenblum} [J. Math. Anal. Appl. 23, 139--147 (1968; Zbl 0159.43102)] and defines a conditional expectation to show the following interesting result. Theorem 7.3: Let \(F(x) \in M_n(\mathbb{C}[x])\) be a self adjoint matrix polynomial. Then \(F(x)\) is positive semidefinite in every point \(x\in [a,b]\) if and only if \(F=G_1^*G_1+(b-x)(x-a)G_2^*G_2\) for some \(G_1,G_2 \in M_n(\mathbb{C}[x]).\) Similar results are shown for half open intervals. Section 8 is concerned with positivstellensätze for \(M_n({\mathsf L})\) with \({\mathsf L}\) as in section 2. If \(B=\text{diag}(1,\lambda_1,\dots, \lambda_{n-1})\) with \(\lambda_i\in \mathbb{R}^\circ\), then for \(X\in M_n({\mathsf L})\) one defines \(X^\tau = B^{-1} X^* B.\) This yields an involution which is associated to the inner product \(\langle x,y \rangle_1 :=\langle Bx,y \rangle,\) where the right hand side uses the standard inner product. Write \(P_B({\mathsf L})\) for the preordering generated by the \(\lambda_i.\) An ordering \(p\) of \({\mathsf R}\) that contains \(\lambda_1, \dots, \lambda_{n-1}\) is a \(*\)-ordering. The positivity is defined by \((M_n({\mathsf L}), \tau)_+\) being the set of matrices \(X\) for which \(X=X^\tau\) and all the principal minors of \(X\) belonging to \(P_B({\mathsf L}).\) It is shown that a type IV positivstellensatz holds for \((M_n({\mathsf L}),\tau).\) \textit{I. Klep} and \textit{T. Unger} have shown that a type I satz cannot hold for \(n\geq 3;\) see [J. Algebra 324, No. 2, 256--268 (2010; Zbl 1262.16039)]. Section 9 uses results and notation of the previous section to show positivstellensätze for cyclic algebras. Let \({\mathsf L}\) be as before. Let \( {\mathsf L/K} \) be a Galois extension with group \(\mathbb{Z}/(n),\) and \(\mathfrak{A}\) be the cyclic algebra associated to \({\mathsf L/K} \) (see Pierce [loc. cit., Ch. XV]). Positivity is defined by \(\mathfrak{A}_+\) being the set of elements \(y=y^*\) for which \(\mathfrak{p}(x^*yx)\in P_B({\mathsf L} )\) for all \(x\in \mathfrak{A},\) where \(\mathfrak{p}:\mathfrak{A} \rightarrow {\mathsf L} \) is a certain canonical projection. A type IV positivstellensatz holds for \(\mathfrak{A}.\) Section 10 illustrates the results of the previous sections having in common that \(\mathcal{A}_+=\{x\in \mathcal{A}_h: \pi(x) \geq 0 \text{ for all finite dimensional }*\)-representations